A new method on finding optimal centers for improving K-means algorithm
Jie Zhang, Jianrui Dong, Yiyong Xiao · 2015
The mean center (geometric center) has always been used to represent the cluster center in classical K-means algorithm which may cause error. In this paper, a new method, P-partition Method, is introduced to obtain the center of a cluster, which proved efficient and globally optimal. Then two related clustering algorithms are presented by replacing the mean center of K-means based on the new centers found by P-partition, both of which are verified in experiments to be able to provide a better objective value (averagely about 3% lower than that of K-means) under the same conditions.